Skip to main content

Meshroom

TOOL
📖6 min read
Latest Version: 2025.1.0
Photoscanning 100% Free: Complete Meshroom Tutorial | Photogrammetry Course by Gleb Alexandrov

Node Pipeline

Every stage of the reconstruction is a node with its own output. You can inspect intermediate results and switch stages off.

GPU Tuned

The dense stages are built for NVIDIA CUDA GPUs. Other hardware runs the pipeline with reduced performance.

Standard Outputs

Cameras, sparse and dense clouds, and the mesh export as open formats (MCM, PLY, OBJ). Nothing is locked to the application.

Introduction​

Meshroom is a free, open-source photogrammetry pipeline from AliceVision. It runs the same stages as any SfM/MVS pipeline (feature extraction, matching, structure from motion, dense cloud, mesh, texture) as a node graph you can see and edit. Where COLMAP is the command-line tool that makes a run reproducible by design, Meshroom is the GUI that lets you watch every stage, reject a bad frame, and re-run from the node that failed.

Pick Meshroom when you want to iterate visually on a photo set. Pick COLMAP when the reconstruction itself is the evidence and the command chain is what you document. The method is documented in the photogrammetry technique.

Installation​

  1. Download Meshroom from the AliceVision site. Installers exist for Windows and Linux. The macOS build is distributed from the GitHub releases page.
  2. Install the GPU driver that the pipeline expects: an NVIDIA driver with CUDA support. Without it the dense stages run slow or not at all.
  3. Verify the install by opening Meshroom and reading the version in the About dialog. Record that version in your method section.

Usage​

A reconstruction is a project file. You load images, check them, and run the node chain. Each node writes its own output to the project's output folder.

Create the project and load images· .mcm project file

Create a new project and add the image set. What goes in: photos or extracted frames in one folder. What comes out: a project file (.mcm) that stores the node graph and every parameter. The project file is the record of the run, so give it a case filename.

Run image quality check· the QC gate

Run the Image Quality node before anything else. It flags blurred or overexposed frames. Remove the flagged frames from the set. For footage-derived frame sets this step is your quality gate: a bad frame here becomes false geometry later.

Run structure from motion· cameras and sparse cloud

Run the SfM node (SfMNode). It computes every camera position and the sparse point cloud. Open the 3D view: the camera frustums should form a clean shell around the scene, and the sparse cloud should cover the subject. Gaps in the frustums are gaps in your coverage.

Run dense reconstruction and meshing· dense cloud, mesh, texture

Run the Dense Mesh node. It produces the dense point cloud, the textured mesh, and the texture maps. This stage takes the longest and uses the most GPU memory. Lower the output size if the scene exceeds your GPU.

Export· OBJ and PLY

Take the outputs from the project's output folder: the textured mesh (OBJ with texture maps) and the point clouds (PLY). What comes out: files for Blender, and the .mcm project file for the case record.

In an Investigation​

Use the block below to document how you ran Meshroom on the evidence.

Documentation: how to document Meshroom

In Your Method Section​

Guidance: Describe the purpose, the tool version, and the input data.

  • Investigative Objective: To reconstruct the 3D geometry of the scene from a set of 2D images for spatial analysis.
  • Software Version: Meshroom [Version] (e.g., 2025.1.0).
  • Input Data: [Image set name/ID], [N] images derived from [source footage/document], Source Hash: [First 6 digits].

In Your Decision Log​

Guidance: Record the specific procedural parameters and integrity checks.

  • Specific Settings: Frames removed by the image quality check (count and reason), SfM and Dense Mesh node parameters that differ from defaults, GPU used.
  • Assumptions: The scene was static during capture and the camera intrinsics stayed constant.
  • Integrity Check: Analysis ran on a working copy, never the original evidence file.
  • Date Performed: [YYYY-MM-DD].

Verification & Mitigation​

Guidance: How did you verify accuracy and mitigate bias?

  • Verification: Inspected the SfM 3D view: camera frustums form a continuous shell, sparse cloud covers the subject. Compared the reconstructed scale against a known reference dimension.
  • Bias Mitigation: Rejected blurred frames at the image quality stage instead of letting SfM drop them silently.

Example Methodology Statement​

"Scene geometry was reconstructed with Meshroom (2025.1.0) from [N] images of [source] (Hash: A1B2C3). [N] frames were rejected at the image quality stage. Structure from motion registered [N] cameras; the dense mesh was generated at [size] px. Scale was anchored to a reference measurement of [X.X m] and verified against [reference]. The project file [name.mcm] records all node parameters."

Into Blender​

Meshroom outputs a textured mesh and point clouds.

  • Textured mesh (OBJ + texture maps): in Blender, enable the bundled Wavefront OBJ import add-on (Edit > Preferences > Add-ons, search "OBJ") and use File > Import > Wavefront (.obj), or export GLB from an intermediate tool and use File > Import > glTF 2.0 (.glb/.gltf).
  • Point clouds (PLY): import with File > Import > Stanford PLY (.ply), or use the Photogrammetry Importer add-on for georeferenced placement.

Three caveats before you model against the result:

  • Scale is arbitrary. Meshroom has no units, the same as COLMAP. Anchor it to a known dimension before or right after the import.
  • Check the orientation of the imported object and rotate it onto the Blender ground plane.
  • For georeferenced scenes: place the reconstruction relative to the site with ground control points or the QGIS baseline (custom CRS and origin point).

Limitations​

  • The node GUI is harder to script than the COLMAP command line. For fully automated reruns, COLMAP is the better tool.
  • The dense stages assume an NVIDIA CUDA GPU. Other hardware runs with reduced performance.
  • The same SfM weaknesses apply: reflective, transparent, and featureless surfaces, and moving objects, produce holes and false geometry.
  • The output has no real-world scale without a reference dimension or ground control points.

Resources​

Sources​

  1. AliceVision, official site with downloads
  2. Meshroom releases on GitHub, current stable 2025.1.0 (2025-08-18)
  3. Meshroom repository, source, issue tracker, and documentation links